Abstract
We consider discrete-time approximations for path-dependent Isaacs partial differential equations (PDEs) of deterministic differential games under quadratic growth conditions including linear/quadratic problems with distributed and discrete delays. Owing to the path-dependence of the system, the Isaacs PDEs are defined on infinite-dimensional spaces of past state trajectories. Using the notion of viscosity solutions on the infinite-dimensional spaces as proposed by Lukoyanov based on co-invariant derivatives of path spaces, we show that the discrete-time path-dependent dynamic games converge to a unique viscosity solution for the Isaacs PDEs. Noting that these games can be practically defined on finite-dimensional spaces, we discuss finite-dimensional approximations of viscosity solutions of path-dependent Isaacs PDEs. Given an example, we derive discrete-time Riccati-type recursive equations to calculate explicit discrete-time approximations for the path-dependent linear/quadratic problems.
Similar content being viewed by others
References
Bardi, M., Capuzzo-Dolcetta, I.: Optimal Control and Viscosity Solutions of Hamilton–Jacobi–Bellman Equations. Birkhäuser, Boston (1997)
Barles, G., Souganidis, P.E.: Convergence of approximation schemes for fully nonlinear second order equations. Asymptot. Anal. 4, 271–283 (1991)
Bensoussan, A., Nagai, H.: Min–max characterization of a small noise limit on risk-sensitive control. SIAM J. Control Optim. 35(4), 1093–1115 (1997)
Bensoussan, A., Da Prato, G., Delfour, M.C., Mitter, S.K.: Representation and Control of Infinite Dimensional Systems, 2nd edn. Birkhäuser, Boston (2007)
Da Lio, F., Ley, O.: Uniqueness results for second-order Bellman–Isaacs equations under quadratic growth assumptions and applications. SIAM J. Control Optim. 45(1), 74–106 (2006)
Dupire, B.: Functional Itô Calculus, Blooming Portfolio Research Paper, No. 2009-04-FRONTIERS (2009)
Ekren, I., Keller, C., Touzi, N., Zhang, J.: On viscosity solutions of path dependent PDEs. Ann. Probab. 42(1), 204–236 (2014)
Ekren, I., Touzi, N., Zhang, J.: Viscosity solutions of fully nonlinear parabolic path dependent PDEs: Part I. Ann. Probab. 44(2), 1212–1253 (2016)
Ekren, I., Touzi, N., Zhang, J.: Viscosity solutions of fully nonlinear parabolic path dependent PDEs: Part II. Ann. Probab. 44(4), 2507–2553 (2016)
Evans, L.C., Souganidis, P.E.: Differential games and representation formulas of Hamilton–Jacobi–Isaacs equations. Indiana Univ. Math. J. 33(5), 773–797 (1984)
Fabbri, G., Gozzi, F., Swiech, A.: Stochastic Optimal Control in Infinite Dimension. Springer, Cham (2017)
Fleming, W.H.: The convergence problem for differential games. J. Math. Anal. Appl. 3, 102–116 (1961)
Fleming, W.H., Soner, M.H.: Controlled Markov Processes and Viscosity Solutions, 2nd edn. Springer, New York (2006)
Friedman, A.: Differential Games. Wiley, New York (1971)
Gomoyunov, M.I., Lukoyanov, N. Yu., Plaksin, A.R.: Approximation of minimax solutions of Hamilton–Jacobi functional equations for time-delay systems. Proc. Steklov Inst. Math. 1(Suppl. 1), 68–75 (2019)
Isaacs, R.: Differential Games. Wiley, New York (1965)
Kaise, H.: Path-dependent differential games of inf–sup type and Isaacs partial differential equations. In: IEEE 54th Annual Conference on Decision and Control, pp. 1972–1977 (2015)
Kaise, H., Kato, T., Takahashi, Y.: Hamilton–Jacobi partial differential equations with path-dependent terminal cost under superlinear Lagrangian. In: 23rd Mathematical Theory of Networks and Systems, pp. 692–699 (2018)
Kaise, H., McEneaney, W.M.: Idempotent expansions for continuous-time stochastic control. SIAM J. Control Optim. 54(1), 73–98 (2016)
Kim, A.V.: Functional Differential Equations: Applications of i-Smooth Calculus. Kluwer Academic Publishers, Dordrecht (1999)
Krasovskii, N.N., Lukoyanov, N. Yu.: Equations of Hamilton–Jacobi type in hereditary systems: minimax solutions. Proc. Steklov Inst. Math. Suppl. 1, 136–153 (2000)
Lukoyanov, N. Yu.: Functional Hamilton–Jacobi type equations with ci-derivatives in control problems with hereditary information. Nonlinear Funct. Anal. Appl. 8(4), 534–555 (2003)
Lukoyanov, N.Yu.: Approximation of the Hamilton–Jacobi functional equations in systems with hereditary information. In: Proceedings of International Seminar on Control Theory and Theory of Generalized Solutions of Hamilton–Jacobi Equations, Dedicated to the 60th Birthday of Academician A.I. Subbotin, 22–26 June 2005, Ekaterinburg, Russia, vol. 1, pp. 108–115. Ural State University Publishers, Ekaterinburg (2006)
Lukoyanov, N. Yu.: On viscosity solution of functional Hamilton–Jacobi type equations for hereditary systems. Proc. Steklov Inst. Math. Suppl. 2, 190–200 (2007)
Lukoyanov, N. Yu.: Viscosity solution of nonanticipating Hamilton–Jacobi equations. Differ. Equ. 43(12), 1715–1723 (2007)
McEneaney, W.M.: Max-Plus Methods for Nonlinear Control and Estimation. Birkhäuser, Boston (2006)
McEneaney, W.M.: Idempotent method for dynamic games and complexity reduction in min–max expansions. In: Proceedings of the 48th IEEE Conference on Decision and Control, pp. 163–168 (2009)
Peng, S.: Note on viscosity solution of path-dependent PDE and \(G\)-martingale-2nd version (2012). arXiv:1106.1144v2
Pham, T., Zhang, J.: Two person zero-sum game in weak formulation and path-dependent Bellman–Isaacs equation. SIAM J. Control Optim. 52(4), 2090–2121 (2014)
Ren, Z., Tan, X.: On the convergence of monotone schemes for path-dependent PDE. Stoch. Process. Appl. 127, 1738–1762 (2017)
Ren, Z., Touzi, N., Zhang, J.: Comparison of viscosity solutions of fully nonlinear degenerate parabolic path-dependent PDEs. SIAM J. Math. Anal. 49(4), 4093–4116 (2017)
Souganidis, P.E.: Max–min representations and product formulas for the viscosity solutions of Hamilton–Jacobi equations with applications to differential games. Nonlinear Anal. 9(3), 217–257 (1985)
Tang, S., Zhang, F.: Path-dependent optimal stochastic control and viscosity solution of associated Bellman equations. Discrete Contin. Dyn. Syst. 35(11), 5521–5553 (2015)
Zhang, J., Zhuo, J.: Monotone schemes for fully nonlinear parabolic path dependent PDEs. J. Financ. Eng. 1(1), 1–23 (2014)
Acknowledgements
The author would like to thank anonymous referees for their careful readings of the manuscript and helpful comments, which helped him to greatly improve the paper. This work is partially supported by JSPS KAKENHI Grant Numbers 17K05362 and 20K03733.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Proofs of Uniform Estimates on \(\{ V^{(N)} \}_{N=1}^\infty \)
Proofs of Uniform Estimates on \(\{ V^{(N)} \}_{N=1}^\infty \)
Proof of Proposition 3.3
Let \(t_i \le t<t_{i+1}\) and \(x_t \in {\mathbf {X}}_{t}\). Suppose there exists \({\hat{K}}_{i+1} \ge 0\) such that
By (3.2) with (A3),
Noting that \(V^{(N)}(T,x_T) = \varPhi (x_T) \ge -K_\varPhi \), we have the recursive equation of \({\hat{K}}_i\):
Solving the above recursive equations, we have \({\hat{K}}_0=K_\varPhi +K_lT\). Hence we obtain
Let \(t_i \le t <t_{i+1}\) and \(x_t \in {\mathbf {X}}_t\). We suppose that there exists \({\bar{K}}_{i+1} \ge 0\) such that
Taking \(a=0\), we have from (3.2)
Thus, we have
where we suppose \(\varDelta _N \le 1\). Recalling (A4), we have the recursive equation of \({\bar{K}}_i\):
Solving the above recursive equation, we have
Noting that \(1+x \le e^x\) for \(x \ge 0\), we have
\(\square \)
Lemma A.1
Let \(0<\epsilon <1,\) \(t_i \le t < t_{i+1}\) and \(x_t \in {\mathbf {X}}_{t}\). For given \(b, b_{k} \in B \ (k=i+1,i+2,\ldots ,N-1),\) let \(a^{\epsilon }, a^{\epsilon }_{k} \in A \ (k=i+1,i+2,\ldots , N-1)\) be given recursively in the following way:
where \(x^\epsilon _{t_k} \ (k=i+1,i+2,\ldots ,N)\) are given by
Then there exists \(C>0\) depending only on \(K_f, K_\varPhi , K_l, T, \nu \) such that
Proof
Adding the above inequalities, we have
Using Proposition 3.3, (A3) and (A4), we have
\(\square \)
Proof of Proposition 3.4
If \(t=T\), (3.7) holds with \(L=L_\varPhi \) because of (A4). Consider the case where \(t_i \le t< t_{i+1}\). We may suppose that \(V^{(N)}(t,y_t) \ge V^{(N)}(t,x_t)\) without loss of generality. Let \(0<\epsilon <1\). Let \(k=i,i+1,\ldots , N-1\) and \(x^\epsilon _{t_k}\) and \(y^{\epsilon }_{t_k}\) be given. Here, we abuse the notation \(t=t_i\). Take \(a^\epsilon _{k} \in A\) and then \(b^\epsilon _k \in B\) such that
Then, we have
where \( x^\epsilon _{t_{k+1}} =\xi ^{(t_k, x^\epsilon _{t_k}),a^\epsilon _{k},b^\epsilon _{k}}_{t_{k+1}}, \ y^\epsilon _{t_{k+1}} =\xi ^{(t_k, y^\epsilon _{t_k}),a^\epsilon _{k},b^\epsilon _{k}}_{t_{k+1}}. \) Thus we can obtain
We will show that there exists \(C>0\) such that for \(k=i,i+1,\ldots , N\),
We denote by C and \(C_j\) (\(j=1,2,\ldots \)) constants depending on \(L_f, K_f, L_l, K_l, \nu , L_\varPhi , K_\varPhi , T\) for the rest of the present proof. We will prove only the estimate of \(\Vert y^\epsilon _{t_k}\Vert _\infty \) in (A.6) because that of \(\Vert x^\epsilon _{t_k}\Vert _\infty \) can be proved similarly. By using (A2), we have
where
Using (A.7) recursively, we have
Here, we use \(p \ge 1\) in the last inequality. Since \(p^N=(1+K_f\varDelta _N)^N \le e^{K_fT}\) because \(1+x \le e^x\) for \(x \ge 0\), we have
By Cauchy–Schwarz inequality and Lemma A.1, we have
Hence we obtain the second inequality of (A.6).
Estimating the RHS of (A.5) with (A.6), we have
Noting that
and \((1+L_f \varDelta _N)^N \le e^{L_fT}\), we have
By (A.9), we have
Letting \(\epsilon \rightarrow 0\), we obtain
\(\square \)
Proof of Proposition 3.5
Note that we can use Lemma 4.3 for \(\varphi =V^{(N)}\) because Proposition 3.4 holds. We suppose N is sufficiently large such that Lemma 4.3 holds for \(\varphi =V^{(N)}\). More precisely, suppose \(h_L > T/N\). We denote by \(C_i\) \((i=1,2,\ldots )\) constants depending on only \(L_f\), \(M_f\), \(K_f\), \(L_l\), \(M_l\), \(K_l\), \(\nu \), \(L_\varPhi \), \(K_\varPhi \), and T in the present proof.
Suppose \(t_i \le t< s < t_{i+1}\) and \(x_t \in {\mathbf {X}}_t\). We consider the case where \(V^{(N)}(s,x_s(\cdot \wedge t)) \ge V^{(N)}(t,x_t)\). We omit the proof for the case where \(V^{(N)}(s,x_s(\cdot \wedge t)) \le V^{(N)}(t,x_t)\) because it can be proved in a manner similar to the above case. Take \({a}^*=a^*(t,x_t;t_{i+1}-t)\) such that
We note that from Lemma 4.3 with Proposition 3.4
where \(C_1={\hat{C}}_L\) with L of Proposition 3.4 which depends on \(K_f\), \(K_l\), \(K_\varPhi \), \(L_f\), \(L_l\), \(L_\varPhi \), \(\nu \), T. Using the definitions of \(V^{(N)}(s,x_s(\cdot \wedge t))\) and \(V^{(N)}(t,x_t)\), we have
By (A3) and (A.12), we have
Similarly, by (A3) and (A.12), we have
Using Proposition 3.4, we have
By the definitions of \(\xi ^{(s,x_s(\cdot \wedge t)),{a}^*,b}_{t_{i+1}}\) and \(\xi ^{(t,x_t),{a}^*,b}_{t_{i+1}}\) with (A2) and (A.12), we have
Using the definitions of \(\xi ^{(s,x_s(\cdot \wedge t)),{a}^*,b}_{t_{i+1}}\) and \(\xi ^{(t,x_t),{a}^*,b}_{t_{i+1}}\) with (A2) and (A.12), we have
Thus, there exists a constant \(C_6>0\) such that
where \(\omega ^{t,x_t}=\omega ^{t,x_t}_f+\omega ^{t,x_t}_l\).
We suppose \(t_i \le t <s=t_{i+1}\). Since the arguments to obtain (A.16) are valid for \(s=t_{i+1}\), we have
From the definition of \(V^{(N)}(t,x_t)\), we have
from which we have
If we take \(a=0\), we have
By (A3), we have
Using Proposition 3.4 and (A2), we have
Thus, we obtain
Hence, we have
Suppose \(0 \le t_i \le t<t_{i+1} \le t_j < s \le t_{j+1}\). Note that
Let \(t_j<s<t_{j+1}\). Applying the first case of (3.8) (resp. the second case of (3.8)) to the first term of (A.18) (resp. the second and third terms of (A.18)), we have
Let \(t_j<s =t_{j+1}\). Applying the second case of (3.8) to each term of (A.18), we obtain
\(\square \)
Rights and permissions
About this article
Cite this article
Kaise, H. Convergence of Discrete-Time Deterministic Games to Path-Dependent Isaacs Partial Differential Equations Under Quadratic Growth Conditions. Appl Math Optim 86, 13 (2022). https://doi.org/10.1007/s00245-022-09829-4
Accepted:
Published:
DOI: https://doi.org/10.1007/s00245-022-09829-4